Catálogo de publicaciones - libros

Color is an essential feature that describes the image content and therefore colors occurring in the images should be effectively characterized in image classification. The selection of the number of the quantization levels is an important matter in the color description. On the other hand, when color representations using different quantization levels are combined, more accurate multilevel color description can be achieved. In this paper, we present a novel approach to multilevel color description of natural rock images. The description is obtained by combining separate base classifiers that use image histograms at different quantization levels as their inputs. The base classifiers are combined using classification probability vector (CPV) method that has proved to be an accurate way of combining classifiers in image classification.

We present in this paper a new non-parametric method for polygonal approximations of digital curves. In classical polygonal approximation algorithms, a starting point is randomly chosen on the curve and heuristics are used to ensure its effectiveness. We propose to use a new canonical representation of digital curves where no point is privileged. We restrict the class of approximation polygons to the class of digital polygonalizations of the curve. We describe the first algorithm which computes the polygon with minimal Integral Summed Squared Error in the class in both linear time and space, which is optimal, independently of any starting point.

We present a novel method for manifold learning, i.e. identification of the low-dimensional manifold-like structure present in a set of data points in a possibly high-dimensional space. The main idea is derived from the concept of Riemannian normal coordinates. This coordinate system is in a way a generalization of Cartesian coordinates in Euclidean space. We translate this idea to a cloud of data points in order to perform dimension reduction. Our implementation currently uses Dijkstra’s algorithm for shortest paths in graphs and some basic concepts from differential geometry. We expect this approach to open up new possibilities for analysis of e.g. shape in medical imaging and signal processing of manifold-valued signals, where the coordinate system is “learned” from experimental high-dimensional data rather than defined analytically using e.g. models based on Lie-groups.

The problem of constructing a tight isothetic outer (or inner) polygon covering an arbitrarily shaped 2D object on a background grid, is addressed in this paper, and a novel algorithm is proposed. Such covers have many applications to image mining, rough sets, computational geometry, and robotics. Designing efficient algorithms for these cover problems was an open problem in the literature. The elegance of the proposed algorithm lies in utilizing the inherent combinatoral properties of the relative arrangement of the object and the grid lines. The shape and the relative error of the polygonal cover can be controlled by changing the granularity of the grid. Experimental results on various complex objects with variable grid sizes have been reported to demonstrate the versatility, correctness, and speed of the algorithm.

We discuss the connection between Gabor filters and steerable filters in pattern recognition. We derive optimal steering coefficients for Gabor filters and evaluate the accuracy of the approximative orientation steering numerically. Gabor filters can be well steerable, but the error of the approximation depends heavily on the parameters. We show how a rotation invariant feature similarity measure can be obtained using steerability.

The problem addressed in nonlinear dimensionality reduction, is to find lower dimensional configurations of high dimensional data, thereby revealing underlying structure. One popular method in this regard is the Isomap algorithm, where local information is used to find approximate geodesic distances. From such distance estimations, lower dimensional representations, accurate on a global scale, are obtained by multidimensional scaling. The property of global approximation sets Isomap in contrast to many competing methods, which approximate only locally.

A serious drawback of Isomap is that it is topologically instable, i.e., that incorrectly chosen algorithm parameters or perturbations of data may abruptly alter the resulting configurations. To handle this problem, we propose new methods for more robust approximation of the geodesic distances. This is done using a viewpoint of electric circuits. The robustness is validated by experiments. By such an approach we achieve both the stability of local methods and the global approximation property of global methods.

Images are being produced and made available in ever increasing numbers; but how can we find images “like this one” that are of interest to us? Many different systems have been developed which offer content-based image retrieval (CBIR), using low-level features such as colour, texture and shape; but how can the retrieval performance of such systems be measured? We have produced a perceptually-derived ranking of similar images using the Brodatz textures image dataset, based on a human study, which can be used to benchmark retrieval performance. In this paper, we show how a “mental map” may be derived from individual judgements to provide a scale of psychological distance, and a visual indication of image similarity.

In this study a method for automatic motor condition diagnosis is proposed. The method is based on a statistical discriminance measure which can be used to select the most discriminative features. New signals are classified to either a normal condition class or a failure class. The classification can be done traditionally using training examples from the both classes or using only probability distribution of the normal condition samples. The latter corresponds to typical situations in practice where the amount of failure data is insufficient. The results are verified using real measurements from induction motors in normal condition and with bearing faults.

We present an Outlier Removal Clustering (ORC) algorithm that provides outlier detection and data clustering simultaneously. The method employs both clustering and outlier discovery to improve estimation of the centroids of the generative distribution. The proposed algorithm consists of two stages. The first stage consist of purely K-means process, while the second stage iteratively removes the vectors which are far from their cluster centroids. We provide experimental results on three different synthetic datasets and three map images which were corrupted by lossy compression. The results indicate that the proposed method has a lower error on datasets with overlapping clusters than the competing methods.

Discrete geometric estimators approach geometric quantities on digitized shapes without any knowledge of the continuous shape. A classical yet difficult problem is to show that an estimator asymptotically converges toward the true geometric quantity as the resolution increases. We study here the convergence of local estimators based on Digital Straight Segment (DSS) recognition. It is closely linked to the asymptotic growth of maximal DSS, for which we show bounds both about their number and sizes. These results not only give better insights about digitized curves but indicate that curvature estimators based on local DSS recognition are not likely to converge. We indeed invalidate an hypothesis which was essential in the only known convergence theorem of a discrete curvature estimator. The proof involves results from arithmetic properties of digital lines, digital convexity, combinatorics, continued fractions and random polytopes.